Funded project, 2024 PredicTCR web service
Project Outcome
The SSC project led to the development of a web service, through which users can use an ML classifier that has been developed in the group of Ed Green. This ML model can identify tumor-reactive T cells in patient samples. Through the support of the SSC, users can now use this model without the need of directly accessing it. This has greatly improved the usability, and resolved the requirement of case-by-case negotiations of an MTA to prevent copying of the model. In the first 2.5 months of use nearly 100 external groups applied for access; this generates a lot of value for the research group as each dataset the users provide would have cost over 3000 Euro to produce, and will lead to novel insights over time as the group of Ed Green performs meta analyses on the submitted data. On the web server, users can sign up for an account, log in, upload sequences, and download their results. The web frontend connects to a distributed runner system that provides docker images to do the model inference and uploads the results. Additionally, there is an admin interface where user permissions, quotas, website settings and content can be updated easily by the research group. The software tech stack involves a vue.js frontend, flask backend, REST API, JWT authentication, and docker, and is deployed as a set of docker services for portability, that are updated automatically when the underlying code is updated (CI/CD pipeline).
The SSC helped us make our research results available to academics around the world by building a flexible platform to access our machine learning classifier. These academics are providing us with new data, leading to a cycle of new insights and further improvement of our tools.
Dr. Edward Green, Immunogenomics Research Group, DKFZ
Enabling new research
Without the predicTCR web service, scientists from other research groups and institutions would not be able to benefit from the tools generated in the Ed Green research group. Through the widespread use and adoption, the group of Ed Green benefits in turn, since their research garners citations as it impacts the wider community. Further, the research group itself can refine and further develop their tools with the help of the users through the gained access to these external datasets. Importantly these external datasets - single cell sequencing data from patient tumor sample - are often from patients of different ethnicities to the patients seen in Heidelberg, which adds additional noise useful for training the machine learning models. Already nearly 100 researchers from around the world have applied to use the tool, and the group is already in close contact with several of these researchers with the aim of establishing collaborations.
The flexible solution provided by the SSC is very helpful to the group, as it will allow to connect new tools as they become available. The interface provides a single point of access to all previous samples, making retrospective re-analysis much much easier for the researchers. Moreover, the researchers highlight that the SSC provides great examples of best practice software engineering - where this collaboration has showcased the value of doing things properly.
Resulting publications
The following publications have directly resulted from this project (note that research groups using the web service may further have published results based on the predicTCR analysis):
- European Society for Molecular Oncology (ESMO) - The Molecular Analysis for Precision Oncology Congress 10/2024, International Conference on Lymphocyte engineering 02/2025
